One-Line Summary

Agent-focused infrastructure hits customers’ racks as Apple bakes AI into iOS and enterprises fund data and portability layers to run agents at scale.

Big Tech

Nvidia ships Vera, a CPU built for AI agents

Nvidia delivers the first Vera CPU systems—designed to handle the CPU-side work of AI agents—to Anthropic, OpenAI, SpaceXAI, and Oracle Cloud Infrastructure, moving the chip from announcement to customer hands. Introduced at GTC in March, Vera targets the orchestration, tool-calling, retrieval, and code execution steps that agents rely on. ¹

Vera features 88 custom Olympus cores, 1.2 TB/s memory bandwidth, and claims 50% faster per-core performance, with a design meant to keep GPUs fed and overall response times down for agent workflows. Nvidia positions Vera as the host processor in systems that pair tightly with Rubin GPUs via second‑generation NVLink‑C2C and unified memory. ¹

Oracle Cloud Infrastructure says it plans to deploy “hundreds of thousands” of Vera CPUs beginning in 2026 and is described as the first cloud provider taking the chip to hyperscale, signaling that CPU capacity—not only GPUs—will shape agent performance in production. ¹

Meanwhile, national labs emphasize that scientific computing still needs strong double‑precision performance; Sandia is testing newcomer NextSilicon amid concerns that some AI‑optimized chips underweight these workloads, underscoring the broader shift in compute design priorities. Nvidia says it remains committed to balanced chips for real‑world science and AI. ²

Apple prepares AI writing help and natural-language Shortcuts in iOS 27

Apple is preparing built‑in AI writing tools for iOS 27 and iPadOS 27, including a grammar checker, a “Help Me Write” experience, natural‑language creation of systemwide Shortcuts, and AI‑generated wallpapers—aimed at narrowing feature gaps with rival devices. ³

9to5Mac reports UI details: a translucent panel that shows suggested revisions, a “Write With Siri” toggle atop the keyboard, and a Shortcuts prompt—“What do you want your shortcut to do?”—that builds automations from plain descriptions. Wallpaper generation would appear as an option in the standard picker via Image Playground. ⁴

Industry & Biz

Decart raises $300M with Nvidia backing to ease chip switching

Decart raises $300 million led by Radical Ventures, with Nvidia among backers, valuing the San Francisco startup near $4 billion. Its software aims to let AI developers switch between processors from Nvidia, Amazon, Google and others with less friction. ⁵

The pitch: reduce the complexity and cost of moving workloads across vendors and clouds as compute demand stays red‑hot. Decart previously raised $153 million in August and was founded in 2023 by Dean and Orian Leitersdorf and Moshe Shalev. ⁵

Publicis to buy LiveRamp for $2.2B to power data collaboration

Publicis Groupe agrees to acquire LiveRamp for a total enterprise value of $2.167 billion in an all‑cash deal at $38.50 per share, a 29.8% premium to LiveRamp’s May 15 close. Publicis frames the move as enabling “data co‑creation” to build smarter agents. ⁶

Publicis says the deal supports higher 2027–2028 constant‑currency targets of 7%–8% net revenue growth and 8%–10% headline EPS growth. LiveRamp will continue as a neutral, interoperable platform under CEO Scott Howe within Publicis’ Technology segment. ⁷

New Tools

Dust raises $40M to grow 'multiplayer' enterprise agents

Dust offers a platform where teams and AI agents collaborate with shared context and tools across Slack, Notion, GitHub, and Google Drive. The company raises a $40 million Series B from Abstract, Sequoia, Snowflake Ventures, and Datadog to scale the product. ⁸

Dust reports more than 3,000 organizations using the platform with over 300,000 agents deployed and says it had zero customer churn in 2025, while advocating for “AI Operators” inside functions like Sales and Support to build and run these agents. ⁸

What This Means for You

Vera’s delivery highlights how much agent performance depends on CPUs: tool calls, orchestration, retrieval, and code execution all add up. If your stack relies on cloud providers, procurement and SLOs may hinge on CPU availability and scheduling as much as on GPUs. ¹

If you review or produce content on iPhone or iPad, Apple’s planned grammar checker and “Help Me Write” could streamline approvals, while natural‑language Shortcuts lower the barrier to everyday automation. Start noting repeat edits and routine actions you’d want the system to handle. ³

Rapid funding for portability layers like Decart signals a push to hedge against vendor lock‑in. If your AI workload sits with one chip or cloud today, knowing how to switch—or run in parallel—could become a negotiating lever on price, availability, and performance. ⁵

Publicis’ LiveRamp deal puts data collaboration and clean rooms at the center of agent usefulness. Whether or not you use these vendors, the message is clear: governed ways to connect partner data will determine how reliable and differentiated your AI agents become. ⁷

Action Items

1. Map CPU-heavy agent tasks: List the steps where your agents call tools, search long contexts, or execute code, then ask IT or vendors how CPU scaling and scheduling affect response times.
2. Prepare for iOS 27 writing aids: Draft a style guide and a list of 10 repetitive edits or tones you want automated so you can evaluate Apple’s built‑in grammar and “Help Me Write” features quickly.
3. Describe three Shortcuts in plain language: Write natural‑language briefs for routine actions you want automated (e.g., rename, file, notify) to convert into Shortcuts when iOS 27 ships.
4. Ask about portability: Request a brief from your AI and cloud vendors on running your workloads across multiple chip backends and what it takes to switch without code rewrites.
5. Pilot a team agent workspace: Convene 3–5 colleagues to co‑design one cross‑team agent (e.g., RFP or proposal assembly) in a shared agent platform with clear permissions and a one‑week success metric.